|
|
|
|
 |
Volume 19, Issue 68 (5-2025) |
 |
|
|
|
|
Assessment and Analysis of Land Use Changes Using Landsat Satellite Images and Random Forest Algorithm (Case Study: Khosuyeh Dam Watershed)
|
Zahra Mohammadii    , Esmaeil Soheili * , Yaghoub Niazi , Farid Foroughi |
|
|
|
Abstract: (123 Views) |
Extended Abstract
Introduction
Land use change analysis is of particular importance due to its direct impact on environmental, social cycles, and ecohydrological conditions of watersheds. Watersheds in arid and semi-arid regions, especially dam and agricultural plains, as are as sensitive to environmental changes, require careful management based on scientific data. Remote sensing data is one of the important tools in this field, which is a suitable alternative to traditional and expensive land surveying methods due to its wide coverage, lower cost, and ease of access. Satellite image classification, as one of the main steps in remote sensing data processing, includes a variety of methods, including algorithms such as maximum likelihood, support vector machines, decision trees, neural networks, and random forests. Previous studies have shown that the random forest algorithm has a high ability to classify satellite images and prepare land cover maps. This method has been used to analyze land use changes in different regions of the world and has brought high accuracy in the results. Among the research conducted, we can mention the application of this algorithm in the assessment of urban, agricultural and forest changes, all of which emphasize the high efficiency and accuracy of this method. In this regard, the accurate identification and analysis of land use change patterns can not only help improve management processes, but also serve as a tool for predicting the long-term effects of environmental changes in watersheds.
Methodology
The study area in this study is the Khosuyeh Dam watershed in Darab County, located in the southeast of Fars Province. This area is considered a semi-arid region of the country, and in line with the objectives of this study, land use changes in this area have been examined in two time periods (2001 and 2021). To analyze land use changes in the Khosuyeh Dam watershed in the time period of 2001 and 2021, satellite data from Landsat 7 for 2001 and Landsat 8 for 2021 were selected. Information on precipitation and temperature in the study area has been collected from meteorological stations. This data has been used to analyze climatic conditions and their changes over time. For the classification of satellite data, the random forest algorithm has been proposed as one of the most effective non-parametric methods. Random forest is one of the powerful and popular algorithms in the field of machine learning and data classification, which is very suitable for complex and noisy data. This algorithm is constructed using a set of decision trees randomly and can effectively classify data into different classes. Due to its high ability to deal with unbalanced data and low sensitivity to small changes in the data, random forest has high accuracy for land cover classification. Accordingly, in this study, the random forest algorithm was used to classify satellite images and simulate land use changes in the study area.
Results
To classify Landsat 7 and 8 satellite images in 2001 and 2021, a random forest algorithm with 40 decision trees was used. This algorithm was selected due to its ability to manage large data, roboticity, and high accuracy in multi-class classification problems, especially in the field of remote sensing. The number of trees, 40, was determined based on the principle of stable error rates. Also, considering the multi-band nature of Landsat satellite images and based on previous studies, the square root of the number of spectral bands (in this study, 7 bands) was considered as the basis for selecting the number of variables examined in each tree. Comparative analysis of land use maps in the two study periods showed significant changes in land use types, especially the reduction of wasteland and the increase in urban areas. These changes are the result of a complex interaction of human and natural factors and can have important consequences on water resources, soil, biodiversity, and ecosystems of the region. Overall, the total land area in 2021 has increased significantly compared to 2001. Each of the land use classes has experienced different changes. Some classes such as "cropland" and "buildings" have seen a significant increase in area, while classes such as "barren land" have seen a decrease in area. These changes reflect different development trends in the study area.
Discussion and Conclusion
In this study, with the aim of analyzing land use changes in the Khosuyeh Darab Dam watershed, Landsat 7 and 8 satellite images from 2001 and 2021 were used. In order to classify land use more accurately, the random forest algorithm was used, and the results indicated the optimal performance of this method with an overall accuracy of 89% in 2001 and 91% in 2021. This indicates the high efficiency of the random forest algorithm in separating land uses and preparing accurate land cover maps. A comparative analysis of land use maps of the two time periods studied showed significant changes in land use types, especially the reduction of barren lands and the increase in urban areas. These changes are the result of the complex interaction of human and natural factors and can have important consequences on water resources, soil, biodiversity, and ecosystems of the region. This research, using modern methods of satellite data analysis and machine learning, has provided a suitable platform for a deeper understanding of environmental changes and providing management solutions. The results obtained provide a valuable tool for managers and planners of the region to formulate natural resource conservation and sustainable development programs. In addition, the methodology used in this research has the potential to be generalized to other similar regions in the country. Similar research has also shown that the use of the random forest algorithm can have high accuracy in land classification, including a study in a semi-arid region that reported an overall accuracy of the random forest algorithm of over 90 percent.
|
Article number: 3 |
| Keywords: Darab Plain, Land Use, Machine Learning, Dam Watershed |
|
|
|
Type of Study: Research |
Subject:
Special
Received: 2024/12/13 | Accepted: 2025/01/16 | Published: 2025/06/9
|
|
|
|
|
|
|
| Send email to the article author |
|
|
|
| Add your comments about this article |
|
|
|
|
mohammadii Z, Soheili E, Niazi Y, Foroughi F. Assessment and Analysis of Land Use Changes Using Landsat Satellite Images and Random Forest Algorithm (Case Study: Khosuyeh Dam Watershed). jwmseir 2025; 19 (68) : 3 URL: http://jwmsei.ir/article-1-1185-en.html
|
|
|
|
|
|
|
